Normally when I've thought about this problem in the past I've thought in terms of formal correctness. Code generation or answering questions.
It's much easier to have that argument about LLMs because of the twin dangers of "user may not know better than the LLM to catch the errors" and "humans are by and large not especially good at finding errors in unfamiliar stacks of right-seeming tokens". The cure to both problems is (a) don't use the damn thing but (b) if you absolutely must, don't use it to do anything you don't already know how to do extremely well.
And I'm not sure that advice works in terms of prose! The users (authors who can't be arsed to *write* which is frankly the only *fun* part of the entire discipline so I really don't understand that mindset) aren't the consumers (readers) and I'm not sure I'm feeling bullish on the critical reading skills of most english speakers my age and younger.
Patch Arcana
npub1vn…vj9ar
2026-05-04 23:35:01 UTC
in reply to nevent1q…0xcd
Author Public Key
npub1vnzk9s963d898c0uzqgnz3zwdwasc2cxkktt6y5f6rrh08qk0hlqzvj9arSeen on
wss://relay.momostr.pinkPublished at
2026-05-04 23:35:01 UTCEvent JSON
{
"id": "625c6874f998366b69050a27df788d1ecea2759fa83696fd2dd5e11e7e11f685",
"pubkey": "64c562c0ba8b4e53e1fc101131444e6bbb0c2b06b596bd1289d0c7779c167dfe",
"created_at": 1777937701,
"kind": 1,
"tags": [
[
"e",
"d6a6d68ad4e1755418d4ff9fb572ee85f0a5d8e2547790fa745deb061563066a",
"",
"reply",
"64c562c0ba8b4e53e1fc101131444e6bbb0c2b06b596bd1289d0c7779c167dfe"
],
[
"e",
"646a01f3dd264fdecd7339d54fdb51553be444869f9022cb3d79f9188f934e85",
"",
"root",
"64c562c0ba8b4e53e1fc101131444e6bbb0c2b06b596bd1289d0c7779c167dfe"
],
[
"proxy",
"https://furry.engineer/@patcharcana/116518925198264343",
"web"
],
[
"p",
"64c562c0ba8b4e53e1fc101131444e6bbb0c2b06b596bd1289d0c7779c167dfe"
],
[
"proxy",
"https://furry.engineer/users/patcharcana/statuses/116518925198264343",
"activitypub"
],
[
"L",
"pink.momostr"
],
[
"l",
"pink.momostr.activitypub:https://furry.engineer/users/patcharcana/statuses/116518925198264343",
"pink.momostr"
],
[
"-"
]
],
"content": "Normally when I've thought about this problem in the past I've thought in terms of formal correctness. Code generation or answering questions.\n\nIt's much easier to have that argument about LLMs because of the twin dangers of \"user may not know better than the LLM to catch the errors\" and \"humans are by and large not especially good at finding errors in unfamiliar stacks of right-seeming tokens\". The cure to both problems is (a) don't use the damn thing but (b) if you absolutely must, don't use it to do anything you don't already know how to do extremely well.\n\nAnd I'm not sure that advice works in terms of prose! The users (authors who can't be arsed to *write* which is frankly the only *fun* part of the entire discipline so I really don't understand that mindset) aren't the consumers (readers) and I'm not sure I'm feeling bullish on the critical reading skills of most english speakers my age and younger.",
"sig": "ef4f9c217b4b3f2ad53ee76632314ab578c6963741cf9fed01ffb21d7379b8c83baa91d3bd1a9192286c19b81e448dadcd0051da3075526d68f1bb5a90670e01"
}